AlphaTransit: Learning to Design City-scale Transit Routes

Researchers introduced AlphaTransit, a planning framework for bus route design with delayed simulator feedback.

It combines Monte Carlo Tree Search with a learned policy-value network to guide route extensions.

On a Bloomington transit benchmark, AlphaTransit achieved the best service rates in both mixed and full demand settings.

The method outperformed both reinforcement learning and unguided search, showing promise for city-scale transit planning.